Modeling and realization of \(n\)-dimensional linear discrete systems (Q1389897)

The class of \(n\)-\(D\) discrete-time linear systems described by the AR representation is studied. An algorithm is given which associates to any \(n\)-\(D\) AR representation an ARMA realization of the form \[ (F\Lambda Q+ E)\xi+ Gw= 0,\tag{1} \] where \(F\in K^{g\times h}\), \(\Lambda= \text{diag}\{z_{i_1},z_{i_2},\dots, z_{i_h}\}\), with \(i_j\) in \(\{1,\dots,n\}\) for \(j= 1,\dots, h\), \(Q\in K^{h\times m}\), \(E\in K^{g\times l}\); here \(K\) is a field, \(z_i\) are indeterminates, \(i\in\{1,\dots, n\}\); the internal variable \(\xi\) and the external variable \(w\) are vectors with entries in \(K^{\mathbb{Z}^n}\). It is shown that the ARMA model (1) includes the well-known Roesser, Fornasini-Marchesini and Attasi (singular) models. The minimality of the ARMA model (1) is discussed with respect to the natural partial order on \(\mathbb{Z}^3_+\) in terms of \((g,h,m)\). Necessary conditions for minimality are obtained, namely, if the ARMA model (1) is minimal then the matrix \([FQ]\) has full row rank and \(Q\) has full column rank.